9 WEARABLES · 30 SOURCES

How accurate is your wearable's step count?

On a treadmill almost every tracker is within a few percent. The same wrist can be thousands of steps out over a normal day, and what decides it is how fast you walk and what your arms are doing, not the badge on the strap.

0.9 m/s

Speed below which every device fails

50.3%

Oura's free-living error, +2,124 steps a day

9

Wearables, grouped in 4 evidence tiers

30

Sources listed, 24 with a permanent link

Build your comparison

Compare devices side by side.

Pick 2 or 3 and compare them on the five things that decide whether a step total is trustworthy. The better value in each row is highlighted.

Choose devices to compare

Tap to add or remove · 2 to 3 devices

Garmin

Apple

Samsung

COROS

Fitbit

Polar

Oura

Pixel

WHOOP

These span Tiers 1 to 3. Read the two error rows together: a device can look excellent on a treadmill and still be thousands of steps out over a normal day, which is exactly what happens to the ring.

Get the Kygo app

The ranking, honestly

These span Tiers 1 to 2. Read the two error rows together: a device can look excellent on a treadmill and still be thousands of steps out over a normal day, which is exactly what happens to the ring.

Careful with this one. Pixel Watch has no published step validation, so the blank cells are missing evidence rather than good results. Only compare the devices that actually have numbers.

Four evidence tiers, not a leaderboard.

The "Garmin is 82.6% accurate" figure everyone quotes comes from a consumer aggregation, not a study. What the research supports is grouping, not ranking.

MAPE is the average percentage a count is off, so lower is better. Lab is a treadmill or structured walk; free-living is a normal day against a research-grade reference. Figures are pooled across the models actually studied, which may not be yours.

A lab number is not a day number. Around 5% in controlled walking, above 10% in real life, on every device here.

Model beats brand. Fitbit spans 3.6% to 35.5%; Garmin spans 1% to 15% low. A brand name without a model number tells you nothing.

No data is not good data. Tier 4 means nobody independent has measured them, not that they failed. It cuts both ways: COROS moved up a tier the moment someone tested it.

Nothing on sale today has been validated for steps. The newest device with published data is about four years old: everything from 2025 on, Apple Series 11 through Oura Ring 5 and Pixel Watch 4, is untested. Two models people ask about do not exist at all: no Fitbit Charge 7, no Polar Vantage V4.

Get the free State of Wearable Accuracy report

What actually decides it

It is your speed and your arms, not the brand.

A wrist tracker does not detect footfalls. It detects arm movement that looks like walking, and everything else follows from that.

Normal arm swing, normal pace

Under 5% error

0.9 to 1.8 m/s, arms free

Walk at a normal or brisk pace with your arms swinging and every device on this page is usable. This is also the condition almost every published accuracy figure was measured in, which is why the marketing sounds so good.

Arms occupied, or a slow pace

35 to 95% of steps missed

Stroller, cart, pockets, handrail

Push a pram or a shopping trolley, carry bags, hold a treadmill handrail or walk with your hands in your pockets and the arm swing disappears. The steps still happen; the watch just does not see them. Slow walking does the same thing for a different reason.

It runs both ways. Gesturing, cooking and road vibration while driving all hit roughly walking cadence, so they add steps you never took. Garmin draws the fewest of these complaints; a finger-worn ring the most. Roos 2020 · Kristiansson 2023 · Hong 2024

The two things that decide it

How fast you walk, and where you wear it.

These two move your step count more than any brand choice. Both are things you control, and neither appears on a spec sheet.

Walking speed

Steps actually counted · higher is better

Jogging or running

over 1.8 m/s

95 to 99%

The highest cadence produces the clearest signal of anything you do all day.

Brisk walking

1.3 to 1.8 m/s

over 95%

The sweet spot for a wrist device, and where most validation studies are run.

Normal walking

0.9 to 1.3 m/s

over 90%

Every device on this page is acceptable here. This is the floor of the good zone.

Slow walking

0.5 to 0.9 m/s

50 to 80%

Older adults, clinical and post-surgical populations spend most of their day here.

Shuffling

under 0.5 m/s

under 50%

Most of your steps are simply never recorded, on every brand at every price.

Bars and pills are roughly how many of your real steps get recorded, pooled across devices, not a single study. Published figures are almost always measured in the top three bands. Roos 2020 · Feehan 2020 · Johnston 2021

Where you wear it

Ranked best to worst

Hip or waist

research-grade placement

0.4 to 5% off

The reference standard in the literature, and inconvenient enough that real people stop wearing it.

Ankle or foot

closest to the actual footfall

2 to 6% off

The best option for slow walkers and clinical use. Very few consumer products support it.

Wrist

what almost everyone wears

5 to 25% off

Reads arm swing as a proxy for walking, which is why it both overcounts and undercounts.

Finger

smart rings

10 to 50%+ off

Hand gestures look like walking. Excellent for sleep and HRV, poor for steps.

The bar is the rough share of steps captured; the pill is the published error range. Wrist and hip totals for the same walk differ by about 30% in young adults and close to 50% in older adults. Oner 2022 · Modave 2017

Read the full factor breakdown

In detail

The full breakdown, tap any device.

What independent research found, what the manufacturer says, and what each device is best and weakest for.

Garmin

Tier 1

Wrist

Independent finding

On a treadmill the Vivosmart HR ran at 0.61 to 1.27% MAPE (Feehan 2020), with acceptable error on lawn, gravel, asphalt and tile. In free living it is the closest brand pooled: about 3% over across 59 studies (Paluch 2026). One study dissents, putting the Fenix 6 15% low over mixed daily activity (Rider 2025).

Error

Lab

0.6 to 3.5%

Real life

+3% over

Behaviour

Bias

Slight overcount; undercounts slow walking

Phantom

Low

Manufacturer claim

No accuracy figure published. Garmin's manual says only that the watch "may interpret some repetitive motions, such as washing dishes, folding laundry, or clapping your hands, as steps".

• How it counts Peak detection on a 3-axis accelerometer. It behaves as though it waits for a run of steps, but the "10-step minimum bout" mechanic everyone quotes is forum folklore, not documented by Garmin.
• Best for Daily totals, runners, and mixed outdoor surfaces.
• Weakest for Mixed daily activity, walking below 1.6 km/h, and treadmill handrails.

View Garmin on Amazon

Apple Watch

Tier 1

Wrist

Independent finding

Over 24 hours against an ActivPAL thigh sensor the Series 6 hit 6.4% MAPE at r=0.99, the tightest free-living agreement in that study (Hong 2024). Not uniform though: slow walking pushed error to 23.9%, and the over-40s averaged 10.9% against 4.3% (Choe & Kang 2025).

Error

Lab

0.9 to 3.4%

Real life

+7% over

Behaviour

Bias

Slight overcount

Phantom

Low

Manufacturer claim

Apple publishes no step accuracy figure and describes step counts as an estimate derived from motion data.

• How it counts Machine-learning peak detection through Core Motion, with GPS-calibrated stride length and arm-rotation data where a gyroscope is fitted.
• Best for Free-living daily totals, and consistency across walking speeds.
• Weakest for Slow walking, and older adults with a shorter, less rhythmic gait.

View Apple Watch on Amazon

Samsung Galaxy Watch

Tier 2

Wrist

Independent finding

Third for steps, behind Garmin and Apple Watch. Over 24 hours against an ActivPAL the Galaxy Watch 4 hit 10.5% MAPE at r=0.82, below the Apple Watch at r=0.99 in the same protocol (Hong 2024). On a slow treadmill walk and shuffle the Galaxy Watch Active 2 counted 7.2% low against video (Rowe 2025). No study covers the Galaxy Watch 5 or newer.

Error

Lab

7.2% low

Real life

10.5%

Behaviour

Bias

Undercounts slow walking

Phantom

High

Manufacturer claim

No accuracy figure published. The BioActive sensor is marketed on breadth, not on step error.

• How it counts Accelerometer plus gyroscope through Samsung Health, which can also fuse phone and watch data, a known source of discrepancies.
• Best for Android integration and general smartwatch use.
• Weakest for Desk work, driving and stationary cycling, where counts inflate.

View Galaxy Watch on Amazon

COROS

Tier 2

Wrist

Independent finding

The quiet winner of the only recent hand-tally head-to-head. The Vertix 2 was the sole device of four equivalent to within 10% of criterion across walking, jogging and combined lab steps; Garmin and Polar both missed (Rider 2025). Caveat: one study, twelve people.

Error

Lab

Within 10%, best of 4

Real life

Equivalent in the field

Behaviour

Bias

Slight undercount

Phantom

Low

Manufacturer claim

COROS makes no published step accuracy claim.

• How it counts Peak detection with a continuous-motion filter.
• Best for Mixed daily activity, running cadence and battery life.
• Weakest for Certainty. One small study is real evidence, but not the depth Fitbit, Garmin or Apple carry.

View COROS on Amazon

Fitbit

Tier 3

Wrist

Independent finding

The most-studied brand by a distance, 144 studies in one review alone (Fuller 2020), and the results disagree. Best case: Inspire 2 at 3.6% MAPE (Cheung 2025), Charge 2 within 3.4% of a research reference. Worst case: 17.1 to 35.5% over 24 hours (Giurgiu 2023), and the Sense 18% high (Miwa 2026). Pooled across 59 free-living studies it reads about 15% over (Paluch 2026), which is why it sits below Samsung here.

Error

Lab

3.6 to 8%

Real life

+15% over

Behaviour

Bias

Overcounts in free living

Phantom

Moderate

Manufacturer claim

No accuracy percentage published. The marketing leans on the volume of research rather than a headline error figure.

• How it counts Tri-axial MEMS peak detection plus a machine-learning layer, reworked since the Google acquisition.
• Best for General fitness tracking, on the deepest evidence base of any brand.
• Weakest for Free-living totals. "Fitbit is accurate" means nothing without a model number.

View Fitbit on Amazon

Polar

Tier 3

Wrist

Independent finding

The direction is unresolved, which is worse than a known bias. The A360 was not valid for any walking condition (Roos 2020). Against a hand tally the Grit X ran 17% LOW (Rider 2025); Henriksen 2022 found the opposite, overreporting in free living. The widely quoted "+3.8%" is a magazine test against two phones, with no criterion measure.

Error

Lab

17% low (Grit X)

Real life

Contested

Behaviour

Bias

Contested

Phantom

Moderate

Manufacturer claim

No step accuracy claim. Polar positions these watches on heart rate and training load.

• How it counts Proprietary peak detection on a wrist accelerometer.
• Best for Heart rate and training load, not steps.
• Weakest for Anything where the step total itself matters, in either direction.

View Polar on Amazon

Oura Ring

Tier 3

Finger

Independent finding

Two studies have measured Oura steps against a research reference, and both found the same thing. Over 14 days the Gen 2 ran at 50.3% MAPE, averaging +2,124 steps a day (Kristiansson 2023, corrected). Over a week against an ActiGraph, it overcounted by +1,416 a day, 95% CI 739 to 2,093 (Niela-Vilén 2022). A third put agreement at r=0.77 and concluded Oura cannot replace an ActiGraph for steps (Henriksen 2022). There is no lab step figure anywhere: Kristiansson's lab phase measured calories, not steps.

Error

Lab

No lab step data

Real life

50.3% (+2,124 a day)

Behaviour

Bias

Overcounts

Phantom

Very high

Manufacturer claim

Oura says its May 2025 "Real Steps" update better separates walking from hand movement, and counts fell about 20%. No independent validation of it exists.

• How it counts A machine-learning classifier on a finger-worn accelerometer.
• Best for Sleep, HRV and recovery, which it is genuinely good at.
• Weakest for Steps. Cooking, typing and talking with your hands all read as walking.

View Oura Ring on Amazon

Google Pixel Watch

Tier 4

Wrist

Independent finding

No peer-reviewed validation. What is documented is instability, twice. The Wear OS 5.1 algorithm of March 2025 inflated counts on every model, even off-wrist, and was reverted a month later. A second inflation bug hit in March 2026; Google fixed it but never corrected the history.

Error

Lab

Not published

Real life

Not published

Behaviour

Bias

Overcounted after update

Phantom

High

Manufacturer claim

No accuracy figure published for any version, including the two that shipped and were withdrawn.

• How it counts Wear OS step detection through Fitbit Health Services.
• Best for Google and Fitbit ecosystem integration.
• Weakest for Historical totals. Two inflation bugs in twelve months, and the 2026 one was never backfilled.

WHOOP

Tier 4

Wrist, bicep or body

Independent finding

Steps arrived in an October 2024 firmware update. There is still no peer-reviewed validation of any kind, so nothing here is a measurement.

Error

Lab

Not published

Real life

Not published

Behaviour

Bias

Unknown

Phantom

Unknown

Manufacturer claim

WHOOP calls steps "validated for on-wrist use" but has published no study to support it.

• How it counts Accelerometer cadence plus gyroscope rotation, added by firmware in October 2024.
• Best for Strain and recovery, what the band is built around.
• Weakest for Steps, which WHOOP itself treats as supplementary.

View WHOOP on Amazon

The "View on Amazon" links above are affiliate links. As an Amazon Associate, Kygo Health earns from qualifying purchases, at no extra cost to you.

True for every device

What no brand can fix.

Four things no tracker can fix, because they are about your gait and your hands rather than the sensor.

Walking speed

Below 0.9 m/s, everything fails

Above about 0.9 m/s every device here is usable. Below it accuracy collapses, and under 0.5 m/s even the best miss most of your steps. Anyone recovering from surgery or using a cane spends their day well outside the conditions the published figures were measured in.

Roos 2020 · Feehan 2020 · Johnston 2021

Mobility aids

A rolling walker costs a third of your steps

The largest single error on this page. Across 42 adults aged 51 to 80, an Apple Watch Series 8 undercounted by 36.4% with a rolling walker. Your arm is resting on the frame, not swinging. A waist monitor helps but still misses at very slow speeds.

Özel 2026 (N=42, manual reference)

Gait pathology

Clinical gait is a different problem

Step algorithms are trained on healthy walking. In neurological conditions such as stroke or Parkinson's they detect only 11 to 47% of steps, so a consumer wearable will understate rehabilitation progress, sometimes by more than half.

Sensors 2025 · Johnston 2021

Age and gait

Your gait changes the number

Apple Watch averaged 4.3% error under 40 and 10.9% at 40 and over. Age is not the mechanism: shorter steps, slower speed and reduced arm swing are. Body composition shows the same indirect effect through gait rather than through BMI.

Choe & Kang 2025 · Modave 2017 · Scataglini 2025

Claim vs reality

What gets quoted, and what holds up.

Where the headline figures actually come from, including one this page used to quote itself.

The "82.6% accurate" ranking

“Garmin 82.6%, Apple Watch 81.1%, Fitbit 77.3% overall step accuracy.”

Quoted everywhere, including by an earlier version of this page. It comes from a consumer data aggregation, not a study, and no peer-reviewed paper reports an "overall accuracy percentage" for a step counter. The real measure is MAPE, always per model and per condition.

WellnessPulse 2025 (consumer aggregate, not peer reviewed)

Oura: "Real Steps"

“A new algorithm better distinguishes walking from hand movement.”

Plausible, and counts did fall about 20% after the May 2025 rollout. But a lower number is not a validated number. We searched PubMed, Google Scholar, DOAJ and OpenAIRE, plus a 138-study smart-ring scoping review: nothing validates Real Steps, Ring 4 or Ring 5 for step counting. Every Oura step figure on this page comes from a Gen 2 or Gen 3 ring.

Oura Health 2025 (manufacturer). Literature searched August 2026, no validation found

Google: the enhanced step algorithm

“Wear OS 5.1 improves step detection for strollers, carts and hiking poles.”

It inflated counts on every Pixel Watch and was reverted a month later. A second inflation bug hit in March 2026 and the history was never corrected. The same watch on the same wrist counted differently in three consecutive months.

Google Wear OS release notes (manufacturer)

Polar: "overreports by 3.8%"

“A Vantage M3 logged 10,479 steps against 10,102 and 10,142 on two phones.”

There is no criterion measure in it. A phone in your pocket is not ground truth. The one hand-tally study has a Polar 17% low, the opposite direction. Polar is contested, not simply overreporting.

TechRadar (no criterion) vs Rider 2025 and Henriksen 2022

Garmin: "a 10-step minimum bout filter"

“Garmin records nothing until 10 consecutive steps, then credits all 10.”

Not in any Garmin manual or support document. It traces to a user post on Garmin's community forum from around 2015. The behaviour is real, Garmin does draw the fewest phantom-step complaints, but the mechanism is folklore.

Garmin Venu 4 Owner's Manual

The test that catches all of these. If a source gives an accuracy percentage without naming a criterion measure and an error metric, it is not an accuracy figure.

Beyond the step count

An accurate step count is not an accurate anything else.

Steps, active minutes and calories come off the same sensor on the same wrist, and they are nowhere near equally trustworthy.

Apple Watch Series 6, one study, one set of days

free living

Steps

+2.1%

Active minutes

-46.2%

Calories

+25.9%

The step count was almost perfect. Everything built on top of it was not. Against a research-grade ActiGraph, the same watch over the same days read steps within about 2%, undercounted moderate-to-vigorous activity by nearly half, and overcounted energy expenditure by a quarter. In the same study the Fitbit Sense overcounted steps by 18.0%.

Do not read a good step number as a good day. A footfall is a countable event; active minutes and calories are inferred from intensity, and the inference is where the error lives. Miwa 2026, funded by Chugai Pharmaceutical (authors are employees and shareholders).

The bottom line

If the step total matters to you, wear a Garmin or an Apple Watch, with Samsung third. Fitbit reads about 15% over in daily life. COROS is the surprise, beating both in the one recent hand-tally test. Do not buy an Oura Ring for steps, or judge it by them. Polar is contested, and nobody has measured the Pixel Watch or WHOOP at all.

But no purchase fixes the real limits: nothing is accurate below 0.9 m/s, a walker costs you a third of your steps, where you wear it matters more than what you buy, and an accurate step count still buys you a calorie number that is 26% out. Use it as a trend against your own yesterday, not a measurement against someone else's watch.

Keep exploring

Related tools.

More free, evidence-based tools to get the most out of your wearable.

Reported vs actual

Calorie Burn Accuracy

Enter your reported calorie burn and see the likely real range, with per-activity accuracy.

Wearables · 22 sources

Open

What moves accuracy

Wearable Accuracy: What Actually Affects It

Tattoos, hairy arms, cold hands, strap tightness, which wrist. What the studies found and the free fixes that work tonight.

Wearables · 26 answers

Open

Accuracy vs lab CPET

Garmin

Apple

Polar

Fitbit

Most Accurate VO2 Max Wearable

How accurately do wearables estimate VO2 max vs a lab CPET? 9 devices compared.

Wearables · 13 sources

Open

Want every brand on every metric in one place? Open the wearable accuracy hub

FAQ

Common questions.

Which wearable has the most accurate step count?

Garmin, then Apple Watch, then Samsung. Pooled across 59 free-living studies, Garmin read about 3% over and Apple Watch about 7% over (Paluch 2026). Samsung is third on smaller studies: the Galaxy Watch 4 had 10.5% error over 24 hours, and the Galaxy Watch Active 2 counted 7.2% low on slow walks. Fitbit does not place, reading about 15% over in the same pooled data. The one recent hand-tally head-to-head went to COROS, but it was a single 12-person study.

How accurate is the Apple Watch step counter?

Better than most, and unusually consistent. Over 24 hours against an ActivPAL thigh sensor the Series 6 hit 6.4% error at r=0.99. Two caveats: slow or light walking pushed error to 23.9%, and adults 40 and over averaged 10.9% against 4.3% for under 40. The age effect is really a gait effect.

Is the Oura Ring accurate for step counting?

No, and this is the one device where the research agrees with itself. Every study that has measured Oura steps found it overcounting: +2,124 a day over 14 days against a pedometer, +1,416 a day over a week against a research accelerometer, and a third putting agreement at r=0.77 and concluding Oura cannot replace an ActiGraph for steps. Two different reference devices, same direction, similar size. A finger sensor reads hand movement, so cooking, typing and gesturing all register as walking. Use the ring for sleep and HRV.

What is the biggest factor affecting step count accuracy?

Walking speed, by a distance. Above about 0.9 m/s every device here is usable. Below it accuracy collapses on all of them, and under 0.5 m/s even the best miss most of your steps. Published figures are almost always measured at normal or brisk speeds, while older adults and anyone using a walker routinely walk slower than that.

Does Garmin overcount or undercount steps?

In daily life it slightly overcounts: about 3% over pooled across 59 free-living studies (Paluch 2026), the closest of any brand. It undercounts at slow speeds, and one hand-tally study had the Fenix 6 15% low over mixed activity. Garmin also draws the fewest phantom-step complaints. Be careful with the usual explanation though: the "10-step minimum bout" mechanic is not in any Garmin document, it traces to a user forum post.

Why does my wearable count steps when I am driving or sitting?

Because a wrist device detects arm motion that looks like walking, not footfalls. Road vibration, desk work, brushing your teeth and talking with your hands all match the cadence pattern the algorithm watches for. Garmin says as much in its own manual. Samsung and Oura draw the most reports; Garmin the fewest.

Are step counters accurate if I use a walker, cane or stroller?

No, and this is the largest single error on the page. With a rolling walker an Apple Watch Series 8 undercounted by 36.4% against a manual count, and slow walking cost a further 16.3%. Your arm is resting on the frame instead of swinging, so the watch has nothing to read. A waist-worn monitor helps but still underestimates at very slow speeds.

If my step count is accurate, are my calories accurate too?

No, and the gap is bigger than most people expect. Against a research-grade ActiGraph, the same Apple Watch read steps within 2% but undercounted active minutes by 46% and overcounted calories by 26%. Steps are countable events; active minutes and calories are inferred from intensity, and that is where the error compounds.

Sources

Every claim, traceable.

Every figure checked against the primary record. 30 sources in all, and 24 have a permanent link; the rest are manufacturer posts and consumer tests, listed unlinked rather than dropped. Verified August 2026; rankings updated October 25, 2026.

Meta-analysis

Paluch et al. 2026: Fitbit, Apple and Garmin across 59 free-living studies

J Sport Health Sci 2026:101162. Garmin +3%, Apple +7%, Fitbit +15%

Video count

Rowe et al. 2025: Galaxy Watch Active 2 on slow walks

PLOS ONE 2025. N=27, slow walk and shuffle. -7.2%

Umbrella review

Fuller et al. 2020: 144 Fitbit, 42 Garmin, 28 Apple studies

JMIR mHealth uHealth 2020;8(9):e18694

12 trackers

Roos et al. 2020: 12 consumer wrist trackers

Int J Environ Res Public Health 2020;17(20):7123

Hand tally

Rider et al. 2025: COROS, Garmin, Polar and Suunto

J Meas Phys Behav 2025;8(1). N=12, lab and field

24h free-living

Hong et al. 2024: Apple and Samsung vs ActivPAL

Sensors 2024;24(14):4658. N=104

Oura

Kristiansson et al. 2023: Oura Gen 2 over 14 days

BMC Med Res Methodol 2023;23:50. Use the corrected values

Correction

BMC correction to Kristiansson 2023, 9 Sept 2023

Revises the Table 2 MAPE values

Assistive devices

Özel et al. 2026: walker, cane and crutches

PeerJ 2026;14:e20690. N=42, aged 51 to 80

Derived metrics

Miwa et al. 2026: steps vs active minutes vs calories

PLOS ONE. Chugai-funded; authors are employees

Age meta

Choe & Kang 2025: Apple Watch accuracy by age

Physiol Meas 2025. 56 studies

Garmin

Garmin validity review 2020: five walking surfaces

Int J Environ Res Public Health 2020;17(13):4269

Fitbit review

Feehan et al. 2020: accuracy of Fitbit devices

PeerJ 2020;8:e9381

Fitbit review

Germini et al. 2022: Fitbit across 20 step studies

J Med Internet Res 2022;24(1):e30791

Fitbit

Cheung et al. 2025: Inspire 2 on a treadmill

N=30, manual criterion. MAPE 3.6%, ICC 0.91

Free-living

Giurgiu et al. 2023: Fitbit across 24 hours

Technologies 2023;11(1):29. MAPE 17.1 to 35.5%

Clinical

Straczkiewicz et al. 2023: Fitbit Charge 2

JMIR Cancer 2023;9:e47646. 3.4% difference

Bout detection

Delobelle et al. 2024: Fitbit bout-level detection

Digit Health 2024;10. Sensitivity >87%

Polar and Oura

Henriksen et al. 2022: Polar and Oura vs ActiGraph

JMIR Form Res 2022;6(5):e27248. Oura steps r=0.77

Wear position

Oner et al. 2022: step counting by body placement

Sensors 2022;22(11):3989

Oura

Niela-Vilén et al. 2022: Oura vs ActiGraph over one week

Comput Inform Nurs 2022;40(12):856. N=42. Oura +1,416 steps/day

Demographics

Modave et al. 2017: age, BMI and dominant hand

JMIR mHealth uHealth 2017;5(6):e88

Gait meta

Scataglini et al. 2025: how obesity changes gait

Int J Obes 2025;49(4):541

Method standard

Johnston et al. 2021: INTERLIVE expert statement

Br J Sports Med 2021;55(14):780

Clinical gait

Sensors 2025: step detection in neurological conditions

Only 11 to 47% of steps detected. No permanent identifier

Manufacturer

Oura Health 2025: the "Real Steps" update

May 2025. Counts fell about 20%. Vendor blog, no study

Manufacturer

Google 2025 and 2026: two Wear OS step regressions

Shipped Mar 2025, reverted Apr 2025. Second bug Mar 2026

Manufacturer

Garmin Venu 4 Owner's Manual

The only Garmin text on step accuracy. No bout-filter claim

No criterion

TechRadar: Polar Vantage M3 step test

Source of the "+3.8%". Compared against phones, not a criterion

Consumer aggregate

WellnessPulse and AIM7 2025

Source of the 82.6% / 81.1% / 77.3% figures. Not peer reviewed

Show all 30 sources

From the blog

Show fewer sources

Keep reading.

The long-form, evidence-based articles behind this tool.

Wearables & Data

Which Wearable Counts Steps Most Accurately? 59 Studies Say Garmin (2026)

59 peer-reviewed studies, plus consumer testing for the devices nobody has studied. Every source flagged so you can judge it yourself.

8 min read

Read

Wearables & Data

What Actually Affects Your Wearable's Step Count Accuracy, Ranked by Impact

Walking speed, arm swing and where you wear the device matter more than the brand. Every factor ranked by how much it moves the count.

7 min read

Read

Wearables & Data

What’s the Most Accurate Wearable? 45+ Studies, 7 Brands, Ranked (2026)

45+ independent studies, seven brands, one podium per metric. Which wearable reads closest to the lab for sleep, HRV, heart rate and steps.

10 min read

Read

This content is for informational purposes only and is not medical advice. Consumer step counters are estimates, not clinical measurements, and they are not validated for clinical populations or for gait assessment. Consult a qualified healthcare provider before making health decisions based on wearable data.

Data from peer-reviewed validation studies and manufacturer documentation, with non-peer-reviewed sources labelled as such. Last updated October 25, 2026.

As an Amazon Associate, Kygo Health earns from qualifying purchases. Product links on this page are affiliate links; we may earn a commission at no extra cost to you.